4 * Copyright (C) 1992 Rick Sladkey
6 * nfs directory handling functions
8 * 10 Apr 1996 Added silly rename for unlink --okir
9 * 28 Sep 1996 Improved directory cache --okir
10 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de
11 * Re-implemented silly rename for unlink, newly implemented
12 * silly rename for nfs_rename() following the suggestions
13 * of Olaf Kirch (okir) found in this file.
14 * Following Linus comments on my original hack, this version
15 * depends only on the dcache stuff and doesn't touch the inode
16 * layer (iput() and friends).
17 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM
20 #include <linux/time.h>
21 #include <linux/errno.h>
22 #include <linux/stat.h>
23 #include <linux/fcntl.h>
24 #include <linux/string.h>
25 #include <linux/kernel.h>
26 #include <linux/slab.h>
28 #include <linux/sunrpc/clnt.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/pagemap.h>
32 #include <linux/pagevec.h>
33 #include <linux/namei.h>
34 #include <linux/mount.h>
35 #include <linux/sched.h>
36 #include <linux/kmemleak.h>
37 #include <linux/xattr.h>
39 #include "delegation.h"
44 /* #define NFS_DEBUG_VERBOSE 1 */
46 static int nfs_opendir(struct inode *, struct file *);
47 static int nfs_closedir(struct inode *, struct file *);
48 static int nfs_readdir(struct file *, void *, filldir_t);
49 static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
50 static int nfs_create(struct inode *, struct dentry *, umode_t, struct nameidata *);
51 static int nfs_mkdir(struct inode *, struct dentry *, umode_t);
52 static int nfs_rmdir(struct inode *, struct dentry *);
53 static int nfs_unlink(struct inode *, struct dentry *);
54 static int nfs_symlink(struct inode *, struct dentry *, const char *);
55 static int nfs_link(struct dentry *, struct inode *, struct dentry *);
56 static int nfs_mknod(struct inode *, struct dentry *, umode_t, dev_t);
57 static int nfs_rename(struct inode *, struct dentry *,
58 struct inode *, struct dentry *);
59 static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
60 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
61 static void nfs_readdir_clear_array(struct page*);
63 const struct file_operations nfs_dir_operations = {
64 .llseek = nfs_llseek_dir,
65 .read = generic_read_dir,
66 .readdir = nfs_readdir,
68 .release = nfs_closedir,
69 .fsync = nfs_fsync_dir,
72 const struct inode_operations nfs_dir_inode_operations = {
77 .symlink = nfs_symlink,
82 .permission = nfs_permission,
83 .getattr = nfs_getattr,
84 .setattr = nfs_setattr,
87 const struct address_space_operations nfs_dir_aops = {
88 .freepage = nfs_readdir_clear_array,
92 const struct inode_operations nfs3_dir_inode_operations = {
97 .symlink = nfs_symlink,
101 .rename = nfs_rename,
102 .permission = nfs_permission,
103 .getattr = nfs_getattr,
104 .setattr = nfs_setattr,
105 .listxattr = nfs3_listxattr,
106 .getxattr = nfs3_getxattr,
107 .setxattr = nfs3_setxattr,
108 .removexattr = nfs3_removexattr,
110 #endif /* CONFIG_NFS_V3 */
114 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
115 static int nfs_open_create(struct inode *dir, struct dentry *dentry, umode_t mode, struct nameidata *nd);
116 const struct inode_operations nfs4_dir_inode_operations = {
117 .create = nfs_open_create,
118 .lookup = nfs_atomic_lookup,
120 .unlink = nfs_unlink,
121 .symlink = nfs_symlink,
125 .rename = nfs_rename,
126 .permission = nfs_permission,
127 .getattr = nfs_getattr,
128 .setattr = nfs_setattr,
129 .getxattr = generic_getxattr,
130 .setxattr = generic_setxattr,
131 .listxattr = generic_listxattr,
132 .removexattr = generic_removexattr,
135 #endif /* CONFIG_NFS_V4 */
137 static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, struct rpc_cred *cred)
139 struct nfs_open_dir_context *ctx;
140 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
143 ctx->attr_gencount = NFS_I(dir)->attr_gencount;
146 ctx->cred = get_rpccred(cred);
149 return ERR_PTR(-ENOMEM);
152 static void put_nfs_open_dir_context(struct nfs_open_dir_context *ctx)
154 put_rpccred(ctx->cred);
162 nfs_opendir(struct inode *inode, struct file *filp)
165 struct nfs_open_dir_context *ctx;
166 struct rpc_cred *cred;
168 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
169 filp->f_path.dentry->d_parent->d_name.name,
170 filp->f_path.dentry->d_name.name);
172 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
174 cred = rpc_lookup_cred();
176 return PTR_ERR(cred);
177 ctx = alloc_nfs_open_dir_context(inode, cred);
182 filp->private_data = ctx;
183 if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
184 /* This is a mountpoint, so d_revalidate will never
185 * have been called, so we need to refresh the
186 * inode (for close-open consistency) ourselves.
188 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
196 nfs_closedir(struct inode *inode, struct file *filp)
198 put_nfs_open_dir_context(filp->private_data);
202 struct nfs_cache_array_entry {
206 unsigned char d_type;
209 struct nfs_cache_array {
213 struct nfs_cache_array_entry array[0];
216 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
220 unsigned long page_index;
223 loff_t current_index;
224 decode_dirent_t decode;
226 unsigned long timestamp;
227 unsigned long gencount;
228 unsigned int cache_entry_index;
231 } nfs_readdir_descriptor_t;
234 * The caller is responsible for calling nfs_readdir_release_array(page)
237 struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
241 return ERR_PTR(-EIO);
244 return ERR_PTR(-ENOMEM);
249 void nfs_readdir_release_array(struct page *page)
255 * we are freeing strings created by nfs_add_to_readdir_array()
258 void nfs_readdir_clear_array(struct page *page)
260 struct nfs_cache_array *array;
263 array = kmap_atomic(page);
264 for (i = 0; i < array->size; i++)
265 kfree(array->array[i].string.name);
266 kunmap_atomic(array);
270 * the caller is responsible for freeing qstr.name
271 * when called by nfs_readdir_add_to_array, the strings will be freed in
272 * nfs_clear_readdir_array()
275 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
278 string->name = kmemdup(name, len, GFP_KERNEL);
279 if (string->name == NULL)
282 * Avoid a kmemleak false positive. The pointer to the name is stored
283 * in a page cache page which kmemleak does not scan.
285 kmemleak_not_leak(string->name);
286 string->hash = full_name_hash(name, len);
291 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
293 struct nfs_cache_array *array = nfs_readdir_get_array(page);
294 struct nfs_cache_array_entry *cache_entry;
298 return PTR_ERR(array);
300 cache_entry = &array->array[array->size];
302 /* Check that this entry lies within the page bounds */
304 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
307 cache_entry->cookie = entry->prev_cookie;
308 cache_entry->ino = entry->ino;
309 cache_entry->d_type = entry->d_type;
310 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
313 array->last_cookie = entry->cookie;
316 array->eof_index = array->size;
318 nfs_readdir_release_array(page);
323 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
325 loff_t diff = desc->file->f_pos - desc->current_index;
330 if (diff >= array->size) {
331 if (array->eof_index >= 0)
336 index = (unsigned int)diff;
337 *desc->dir_cookie = array->array[index].cookie;
338 desc->cache_entry_index = index;
346 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
350 int status = -EAGAIN;
352 for (i = 0; i < array->size; i++) {
353 if (array->array[i].cookie == *desc->dir_cookie) {
354 struct nfs_inode *nfsi = NFS_I(desc->file->f_path.dentry->d_inode);
355 struct nfs_open_dir_context *ctx = desc->file->private_data;
357 new_pos = desc->current_index + i;
358 if (ctx->attr_gencount != nfsi->attr_gencount
359 || (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))) {
361 ctx->attr_gencount = nfsi->attr_gencount;
362 } else if (new_pos < desc->file->f_pos) {
364 && ctx->dup_cookie == *desc->dir_cookie) {
365 if (printk_ratelimit()) {
366 pr_notice("NFS: directory %s/%s contains a readdir loop."
367 "Please contact your server vendor. "
368 "The file: %s has duplicate cookie %llu\n",
369 desc->file->f_dentry->d_parent->d_name.name,
370 desc->file->f_dentry->d_name.name,
371 array->array[i].string.name,
377 ctx->dup_cookie = *desc->dir_cookie;
380 desc->file->f_pos = new_pos;
381 desc->cache_entry_index = i;
385 if (array->eof_index >= 0) {
386 status = -EBADCOOKIE;
387 if (*desc->dir_cookie == array->last_cookie)
395 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
397 struct nfs_cache_array *array;
400 array = nfs_readdir_get_array(desc->page);
402 status = PTR_ERR(array);
406 if (*desc->dir_cookie == 0)
407 status = nfs_readdir_search_for_pos(array, desc);
409 status = nfs_readdir_search_for_cookie(array, desc);
411 if (status == -EAGAIN) {
412 desc->last_cookie = array->last_cookie;
413 desc->current_index += array->size;
416 nfs_readdir_release_array(desc->page);
421 /* Fill a page with xdr information before transferring to the cache page */
423 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
424 struct nfs_entry *entry, struct file *file, struct inode *inode)
426 struct nfs_open_dir_context *ctx = file->private_data;
427 struct rpc_cred *cred = ctx->cred;
428 unsigned long timestamp, gencount;
433 gencount = nfs_inc_attr_generation_counter();
434 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
435 NFS_SERVER(inode)->dtsize, desc->plus);
437 /* We requested READDIRPLUS, but the server doesn't grok it */
438 if (error == -ENOTSUPP && desc->plus) {
439 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
440 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
446 desc->timestamp = timestamp;
447 desc->gencount = gencount;
452 static int xdr_decode(nfs_readdir_descriptor_t *desc,
453 struct nfs_entry *entry, struct xdr_stream *xdr)
457 error = desc->decode(xdr, entry, desc->plus);
460 entry->fattr->time_start = desc->timestamp;
461 entry->fattr->gencount = desc->gencount;
466 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
468 if (dentry->d_inode == NULL)
470 if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0)
478 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
480 struct qstr filename = {
484 struct dentry *dentry;
485 struct dentry *alias;
486 struct inode *dir = parent->d_inode;
489 if (filename.name[0] == '.') {
490 if (filename.len == 1)
492 if (filename.len == 2 && filename.name[1] == '.')
495 filename.hash = full_name_hash(filename.name, filename.len);
497 dentry = d_lookup(parent, &filename);
498 if (dentry != NULL) {
499 if (nfs_same_file(dentry, entry)) {
500 nfs_refresh_inode(dentry->d_inode, entry->fattr);
508 dentry = d_alloc(parent, &filename);
512 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
516 alias = d_materialise_unique(dentry, inode);
520 nfs_set_verifier(alias, nfs_save_change_attribute(dir));
523 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
529 /* Perform conversion from xdr to cache array */
531 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
532 struct page **xdr_pages, struct page *page, unsigned int buflen)
534 struct xdr_stream stream;
536 struct page *scratch;
537 struct nfs_cache_array *array;
538 unsigned int count = 0;
541 scratch = alloc_page(GFP_KERNEL);
545 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
546 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
549 status = xdr_decode(desc, entry, &stream);
551 if (status == -EAGAIN)
559 nfs_prime_dcache(desc->file->f_path.dentry, entry);
561 status = nfs_readdir_add_to_array(entry, page);
564 } while (!entry->eof);
566 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
567 array = nfs_readdir_get_array(page);
568 if (!IS_ERR(array)) {
569 array->eof_index = array->size;
571 nfs_readdir_release_array(page);
573 status = PTR_ERR(array);
581 void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
584 for (i = 0; i < npages; i++)
589 void nfs_readdir_free_large_page(void *ptr, struct page **pages,
592 nfs_readdir_free_pagearray(pages, npages);
596 * nfs_readdir_large_page will allocate pages that must be freed with a call
597 * to nfs_readdir_free_large_page
600 int nfs_readdir_large_page(struct page **pages, unsigned int npages)
604 for (i = 0; i < npages; i++) {
605 struct page *page = alloc_page(GFP_KERNEL);
613 nfs_readdir_free_pagearray(pages, i);
618 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
620 struct page *pages[NFS_MAX_READDIR_PAGES];
621 void *pages_ptr = NULL;
622 struct nfs_entry entry;
623 struct file *file = desc->file;
624 struct nfs_cache_array *array;
625 int status = -ENOMEM;
626 unsigned int array_size = ARRAY_SIZE(pages);
628 entry.prev_cookie = 0;
629 entry.cookie = desc->last_cookie;
631 entry.fh = nfs_alloc_fhandle();
632 entry.fattr = nfs_alloc_fattr();
633 entry.server = NFS_SERVER(inode);
634 if (entry.fh == NULL || entry.fattr == NULL)
637 array = nfs_readdir_get_array(page);
639 status = PTR_ERR(array);
642 memset(array, 0, sizeof(struct nfs_cache_array));
643 array->eof_index = -1;
645 status = nfs_readdir_large_page(pages, array_size);
647 goto out_release_array;
650 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
655 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
657 if (status == -ENOSPC)
661 } while (array->eof_index < 0);
663 nfs_readdir_free_large_page(pages_ptr, pages, array_size);
665 nfs_readdir_release_array(page);
667 nfs_free_fattr(entry.fattr);
668 nfs_free_fhandle(entry.fh);
673 * Now we cache directories properly, by converting xdr information
674 * to an array that can be used for lookups later. This results in
675 * fewer cache pages, since we can store more information on each page.
676 * We only need to convert from xdr once so future lookups are much simpler
679 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
681 struct inode *inode = desc->file->f_path.dentry->d_inode;
684 ret = nfs_readdir_xdr_to_array(desc, page, inode);
687 SetPageUptodate(page);
689 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
690 /* Should never happen */
691 nfs_zap_mapping(inode, inode->i_mapping);
701 void cache_page_release(nfs_readdir_descriptor_t *desc)
703 if (!desc->page->mapping)
704 nfs_readdir_clear_array(desc->page);
705 page_cache_release(desc->page);
710 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
712 return read_cache_page(desc->file->f_path.dentry->d_inode->i_mapping,
713 desc->page_index, (filler_t *)nfs_readdir_filler, desc);
717 * Returns 0 if desc->dir_cookie was found on page desc->page_index
720 int find_cache_page(nfs_readdir_descriptor_t *desc)
724 desc->page = get_cache_page(desc);
725 if (IS_ERR(desc->page))
726 return PTR_ERR(desc->page);
728 res = nfs_readdir_search_array(desc);
730 cache_page_release(desc);
734 /* Search for desc->dir_cookie from the beginning of the page cache */
736 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
740 if (desc->page_index == 0) {
741 desc->current_index = 0;
742 desc->last_cookie = 0;
745 res = find_cache_page(desc);
746 } while (res == -EAGAIN);
751 * Once we've found the start of the dirent within a page: fill 'er up...
754 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
757 struct file *file = desc->file;
760 struct nfs_cache_array *array = NULL;
761 struct nfs_open_dir_context *ctx = file->private_data;
763 array = nfs_readdir_get_array(desc->page);
765 res = PTR_ERR(array);
769 for (i = desc->cache_entry_index; i < array->size; i++) {
770 struct nfs_cache_array_entry *ent;
772 ent = &array->array[i];
773 if (filldir(dirent, ent->string.name, ent->string.len,
774 file->f_pos, nfs_compat_user_ino64(ent->ino),
780 if (i < (array->size-1))
781 *desc->dir_cookie = array->array[i+1].cookie;
783 *desc->dir_cookie = array->last_cookie;
787 if (array->eof_index >= 0)
790 nfs_readdir_release_array(desc->page);
792 cache_page_release(desc);
793 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
794 (unsigned long long)*desc->dir_cookie, res);
799 * If we cannot find a cookie in our cache, we suspect that this is
800 * because it points to a deleted file, so we ask the server to return
801 * whatever it thinks is the next entry. We then feed this to filldir.
802 * If all goes well, we should then be able to find our way round the
803 * cache on the next call to readdir_search_pagecache();
805 * NOTE: we cannot add the anonymous page to the pagecache because
806 * the data it contains might not be page aligned. Besides,
807 * we should already have a complete representation of the
808 * directory in the page cache by the time we get here.
811 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
814 struct page *page = NULL;
816 struct inode *inode = desc->file->f_path.dentry->d_inode;
817 struct nfs_open_dir_context *ctx = desc->file->private_data;
819 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
820 (unsigned long long)*desc->dir_cookie);
822 page = alloc_page(GFP_HIGHUSER);
828 desc->page_index = 0;
829 desc->last_cookie = *desc->dir_cookie;
833 status = nfs_readdir_xdr_to_array(desc, page, inode);
837 status = nfs_do_filldir(desc, dirent, filldir);
840 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
844 cache_page_release(desc);
848 /* The file offset position represents the dirent entry number. A
849 last cookie cache takes care of the common case of reading the
852 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
854 struct dentry *dentry = filp->f_path.dentry;
855 struct inode *inode = dentry->d_inode;
856 nfs_readdir_descriptor_t my_desc,
858 struct nfs_open_dir_context *dir_ctx = filp->private_data;
861 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
862 dentry->d_parent->d_name.name, dentry->d_name.name,
863 (long long)filp->f_pos);
864 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
867 * filp->f_pos points to the dirent entry number.
868 * *desc->dir_cookie has the cookie for the next entry. We have
869 * to either find the entry with the appropriate number or
870 * revalidate the cookie.
872 memset(desc, 0, sizeof(*desc));
875 desc->dir_cookie = &dir_ctx->dir_cookie;
876 desc->decode = NFS_PROTO(inode)->decode_dirent;
877 desc->plus = NFS_USE_READDIRPLUS(inode);
878 if (filp->f_pos > 0 && !test_bit(NFS_INO_SEEN_GETATTR, &NFS_I(inode)->flags))
880 clear_bit(NFS_INO_SEEN_GETATTR, &NFS_I(inode)->flags);
882 nfs_block_sillyrename(dentry);
883 res = nfs_revalidate_mapping(inode, filp->f_mapping);
888 res = readdir_search_pagecache(desc);
890 if (res == -EBADCOOKIE) {
892 /* This means either end of directory */
893 if (*desc->dir_cookie && desc->eof == 0) {
894 /* Or that the server has 'lost' a cookie */
895 res = uncached_readdir(desc, dirent, filldir);
901 if (res == -ETOOSMALL && desc->plus) {
902 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
903 nfs_zap_caches(inode);
904 desc->page_index = 0;
912 res = nfs_do_filldir(desc, dirent, filldir);
915 } while (!desc->eof);
917 nfs_unblock_sillyrename(dentry);
920 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
921 dentry->d_parent->d_name.name, dentry->d_name.name,
926 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
928 struct dentry *dentry = filp->f_path.dentry;
929 struct inode *inode = dentry->d_inode;
930 struct nfs_open_dir_context *dir_ctx = filp->private_data;
932 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
933 dentry->d_parent->d_name.name,
937 mutex_lock(&inode->i_mutex);
940 offset += filp->f_pos;
948 if (offset != filp->f_pos) {
949 filp->f_pos = offset;
950 dir_ctx->dir_cookie = 0;
954 mutex_unlock(&inode->i_mutex);
959 * All directory operations under NFS are synchronous, so fsync()
960 * is a dummy operation.
962 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
965 struct dentry *dentry = filp->f_path.dentry;
966 struct inode *inode = dentry->d_inode;
968 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
969 dentry->d_parent->d_name.name, dentry->d_name.name,
972 mutex_lock(&inode->i_mutex);
973 nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
974 mutex_unlock(&inode->i_mutex);
979 * nfs_force_lookup_revalidate - Mark the directory as having changed
980 * @dir - pointer to directory inode
982 * This forces the revalidation code in nfs_lookup_revalidate() to do a
983 * full lookup on all child dentries of 'dir' whenever a change occurs
984 * on the server that might have invalidated our dcache.
986 * The caller should be holding dir->i_lock
988 void nfs_force_lookup_revalidate(struct inode *dir)
990 NFS_I(dir)->cache_change_attribute++;
994 * A check for whether or not the parent directory has changed.
995 * In the case it has, we assume that the dentries are untrustworthy
996 * and may need to be looked up again.
998 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
1000 if (IS_ROOT(dentry))
1002 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1004 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1006 /* Revalidate nfsi->cache_change_attribute before we declare a match */
1007 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1009 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1015 * Return the intent data that applies to this particular path component
1017 * Note that the current set of intents only apply to the very last
1018 * component of the path and none of them is set before that last
1021 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd,
1024 return nd->flags & mask;
1028 * Use intent information to check whether or not we're going to do
1029 * an O_EXCL create using this path component.
1031 static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
1033 if (NFS_PROTO(dir)->version == 2)
1035 return nd && nfs_lookup_check_intent(nd, LOOKUP_EXCL);
1039 * Inode and filehandle revalidation for lookups.
1041 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1042 * or if the intent information indicates that we're about to open this
1043 * particular file and the "nocto" mount flag is not set.
1047 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
1049 struct nfs_server *server = NFS_SERVER(inode);
1051 if (IS_AUTOMOUNT(inode))
1054 /* VFS wants an on-the-wire revalidation */
1055 if (nd->flags & LOOKUP_REVAL)
1057 /* This is an open(2) */
1058 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
1059 !(server->flags & NFS_MOUNT_NOCTO) &&
1060 (S_ISREG(inode->i_mode) ||
1061 S_ISDIR(inode->i_mode)))
1065 return nfs_revalidate_inode(server, inode);
1067 return __nfs_revalidate_inode(server, inode);
1071 * We judge how long we want to trust negative
1072 * dentries by looking at the parent inode mtime.
1074 * If parent mtime has changed, we revalidate, else we wait for a
1075 * period corresponding to the parent's attribute cache timeout value.
1078 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1079 struct nameidata *nd)
1081 /* Don't revalidate a negative dentry if we're creating a new file */
1082 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
1084 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1086 return !nfs_check_verifier(dir, dentry);
1090 * This is called every time the dcache has a lookup hit,
1091 * and we should check whether we can really trust that
1094 * NOTE! The hit can be a negative hit too, don't assume
1097 * If the parent directory is seen to have changed, we throw out the
1098 * cached dentry and do a new lookup.
1100 static int nfs_lookup_revalidate(struct dentry *dentry, struct nameidata *nd)
1103 struct inode *inode;
1104 struct dentry *parent;
1105 struct nfs_fh *fhandle = NULL;
1106 struct nfs_fattr *fattr = NULL;
1109 if (nd->flags & LOOKUP_RCU)
1112 parent = dget_parent(dentry);
1113 dir = parent->d_inode;
1114 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1115 inode = dentry->d_inode;
1118 if (nfs_neg_need_reval(dir, dentry, nd))
1123 if (is_bad_inode(inode)) {
1124 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1125 __func__, dentry->d_parent->d_name.name,
1126 dentry->d_name.name);
1130 if (nfs_have_delegation(inode, FMODE_READ))
1131 goto out_set_verifier;
1133 /* Force a full look up iff the parent directory has changed */
1134 if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
1135 if (nfs_lookup_verify_inode(inode, nd))
1136 goto out_zap_parent;
1140 if (NFS_STALE(inode))
1144 fhandle = nfs_alloc_fhandle();
1145 fattr = nfs_alloc_fattr();
1146 if (fhandle == NULL || fattr == NULL)
1149 error = NFS_PROTO(dir)->lookup(NFS_SERVER(dir)->client, dir, &dentry->d_name, fhandle, fattr);
1152 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1154 if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1157 nfs_free_fattr(fattr);
1158 nfs_free_fhandle(fhandle);
1160 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1163 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
1164 __func__, dentry->d_parent->d_name.name,
1165 dentry->d_name.name);
1168 nfs_zap_caches(dir);
1170 nfs_mark_for_revalidate(dir);
1171 if (inode && S_ISDIR(inode->i_mode)) {
1172 /* Purge readdir caches. */
1173 nfs_zap_caches(inode);
1174 /* If we have submounts, don't unhash ! */
1175 if (have_submounts(dentry))
1177 if (dentry->d_flags & DCACHE_DISCONNECTED)
1179 shrink_dcache_parent(dentry);
1182 nfs_free_fattr(fattr);
1183 nfs_free_fhandle(fhandle);
1185 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
1186 __func__, dentry->d_parent->d_name.name,
1187 dentry->d_name.name);
1190 nfs_free_fattr(fattr);
1191 nfs_free_fhandle(fhandle);
1193 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n",
1194 __func__, dentry->d_parent->d_name.name,
1195 dentry->d_name.name, error);
1200 * This is called from dput() when d_count is going to 0.
1202 static int nfs_dentry_delete(const struct dentry *dentry)
1204 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
1205 dentry->d_parent->d_name.name, dentry->d_name.name,
1208 /* Unhash any dentry with a stale inode */
1209 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1212 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1213 /* Unhash it, so that ->d_iput() would be called */
1216 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1217 /* Unhash it, so that ancestors of killed async unlink
1218 * files will be cleaned up during umount */
1225 static void nfs_drop_nlink(struct inode *inode)
1227 spin_lock(&inode->i_lock);
1228 if (inode->i_nlink > 0)
1230 spin_unlock(&inode->i_lock);
1234 * Called when the dentry loses inode.
1235 * We use it to clean up silly-renamed files.
1237 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1239 if (S_ISDIR(inode->i_mode))
1240 /* drop any readdir cache as it could easily be old */
1241 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1243 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1245 nfs_complete_unlink(dentry, inode);
1250 static void nfs_d_release(struct dentry *dentry)
1252 /* free cached devname value, if it survived that far */
1253 if (unlikely(dentry->d_fsdata)) {
1254 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1257 kfree(dentry->d_fsdata);
1261 const struct dentry_operations nfs_dentry_operations = {
1262 .d_revalidate = nfs_lookup_revalidate,
1263 .d_delete = nfs_dentry_delete,
1264 .d_iput = nfs_dentry_iput,
1265 .d_automount = nfs_d_automount,
1266 .d_release = nfs_d_release,
1269 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
1272 struct dentry *parent;
1273 struct inode *inode = NULL;
1274 struct nfs_fh *fhandle = NULL;
1275 struct nfs_fattr *fattr = NULL;
1278 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
1279 dentry->d_parent->d_name.name, dentry->d_name.name);
1280 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1282 res = ERR_PTR(-ENAMETOOLONG);
1283 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1287 * If we're doing an exclusive create, optimize away the lookup
1288 * but don't hash the dentry.
1290 if (nfs_is_exclusive_create(dir, nd)) {
1291 d_instantiate(dentry, NULL);
1296 res = ERR_PTR(-ENOMEM);
1297 fhandle = nfs_alloc_fhandle();
1298 fattr = nfs_alloc_fattr();
1299 if (fhandle == NULL || fattr == NULL)
1302 parent = dentry->d_parent;
1303 /* Protect against concurrent sillydeletes */
1304 nfs_block_sillyrename(parent);
1305 error = NFS_PROTO(dir)->lookup(NFS_SERVER(dir)->client, dir, &dentry->d_name, fhandle, fattr);
1306 if (error == -ENOENT)
1309 res = ERR_PTR(error);
1310 goto out_unblock_sillyrename;
1312 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1313 res = ERR_CAST(inode);
1315 goto out_unblock_sillyrename;
1318 res = d_materialise_unique(dentry, inode);
1321 goto out_unblock_sillyrename;
1324 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1325 out_unblock_sillyrename:
1326 nfs_unblock_sillyrename(parent);
1328 nfs_free_fattr(fattr);
1329 nfs_free_fhandle(fhandle);
1333 #ifdef CONFIG_NFS_V4
1334 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
1336 const struct dentry_operations nfs4_dentry_operations = {
1337 .d_revalidate = nfs_open_revalidate,
1338 .d_delete = nfs_dentry_delete,
1339 .d_iput = nfs_dentry_iput,
1340 .d_automount = nfs_d_automount,
1341 .d_release = nfs_d_release,
1345 * Use intent information to determine whether we need to substitute
1346 * the NFSv4-style stateful OPEN for the LOOKUP call
1348 static int is_atomic_open(struct nameidata *nd)
1350 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
1352 /* NFS does not (yet) have a stateful open for directories */
1353 if (nd->flags & LOOKUP_DIRECTORY)
1355 /* Are we trying to write to a read only partition? */
1356 if (__mnt_is_readonly(nd->path.mnt) &&
1357 (nd->intent.open.flags & (O_CREAT|O_TRUNC|O_ACCMODE)))
1362 static fmode_t flags_to_mode(int flags)
1364 fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1365 if ((flags & O_ACCMODE) != O_WRONLY)
1367 if ((flags & O_ACCMODE) != O_RDONLY)
1372 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags)
1374 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags));
1377 static int do_open(struct inode *inode, struct file *filp)
1379 nfs_fscache_set_inode_cookie(inode, filp);
1383 static int nfs_intent_set_file(struct nameidata *nd, struct nfs_open_context *ctx)
1388 /* If the open_intent is for execute, we have an extra check to make */
1389 if (ctx->mode & FMODE_EXEC) {
1390 ret = nfs_may_open(ctx->dentry->d_inode,
1392 nd->intent.open.flags);
1396 filp = lookup_instantiate_filp(nd, ctx->dentry, do_open);
1398 ret = PTR_ERR(filp);
1400 nfs_file_set_open_context(filp, ctx);
1402 put_nfs_open_context(ctx);
1406 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1408 struct nfs_open_context *ctx;
1410 struct dentry *res = NULL;
1411 struct inode *inode;
1415 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
1416 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1418 /* Check that we are indeed trying to open this file */
1419 if (!is_atomic_open(nd))
1422 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
1423 res = ERR_PTR(-ENAMETOOLONG);
1427 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1429 if (nd->flags & LOOKUP_EXCL) {
1430 d_instantiate(dentry, NULL);
1434 open_flags = nd->intent.open.flags;
1435 attr.ia_valid = ATTR_OPEN;
1437 ctx = create_nfs_open_context(dentry, open_flags);
1438 res = ERR_CAST(ctx);
1442 if (nd->flags & LOOKUP_CREATE) {
1443 attr.ia_mode = nd->intent.open.create_mode;
1444 attr.ia_valid |= ATTR_MODE;
1445 attr.ia_mode &= ~current_umask();
1447 open_flags &= ~(O_EXCL | O_CREAT);
1449 if (open_flags & O_TRUNC) {
1450 attr.ia_valid |= ATTR_SIZE;
1454 /* Open the file on the server */
1455 nfs_block_sillyrename(dentry->d_parent);
1456 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr);
1457 if (IS_ERR(inode)) {
1458 nfs_unblock_sillyrename(dentry->d_parent);
1459 put_nfs_open_context(ctx);
1460 switch (PTR_ERR(inode)) {
1461 /* Make a negative dentry */
1463 d_add(dentry, NULL);
1466 /* This turned out not to be a regular file */
1471 if (!(nd->intent.open.flags & O_NOFOLLOW))
1475 res = ERR_CAST(inode);
1479 res = d_add_unique(dentry, inode);
1480 nfs_unblock_sillyrename(dentry->d_parent);
1483 ctx->dentry = dget(res);
1486 err = nfs_intent_set_file(nd, ctx);
1490 return ERR_PTR(err);
1493 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1496 return nfs_lookup(dir, dentry, nd);
1499 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1501 struct dentry *parent = NULL;
1502 struct inode *inode;
1504 struct nfs_open_context *ctx;
1506 int openflags, ret = 0;
1508 if (nd->flags & LOOKUP_RCU)
1511 inode = dentry->d_inode;
1512 if (!is_atomic_open(nd) || d_mountpoint(dentry))
1515 parent = dget_parent(dentry);
1516 dir = parent->d_inode;
1518 /* We can't create new files in nfs_open_revalidate(), so we
1519 * optimize away revalidation of negative dentries.
1521 if (inode == NULL) {
1522 if (!nfs_neg_need_reval(dir, dentry, nd))
1527 /* NFS only supports OPEN on regular files */
1528 if (!S_ISREG(inode->i_mode))
1530 openflags = nd->intent.open.flags;
1531 /* We cannot do exclusive creation on a positive dentry */
1532 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1534 /* We can't create new files here */
1535 openflags &= ~(O_CREAT|O_EXCL);
1537 ctx = create_nfs_open_context(dentry, openflags);
1542 attr.ia_valid = ATTR_OPEN;
1543 if (openflags & O_TRUNC) {
1544 attr.ia_valid |= ATTR_SIZE;
1550 * Note: we're not holding inode->i_mutex and so may be racing with
1551 * operations that change the directory. We therefore save the
1552 * change attribute *before* we do the RPC call.
1554 inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, &attr);
1555 if (IS_ERR(inode)) {
1556 ret = PTR_ERR(inode);
1569 if (inode != dentry->d_inode)
1572 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1573 ret = nfs_intent_set_file(nd, ctx);
1583 put_nfs_open_context(ctx);
1589 return nfs_lookup_revalidate(dentry, nd);
1592 static int nfs_open_create(struct inode *dir, struct dentry *dentry,
1593 umode_t mode, struct nameidata *nd)
1595 struct nfs_open_context *ctx = NULL;
1598 int open_flags = O_CREAT|O_EXCL;
1600 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1601 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1603 attr.ia_mode = mode;
1604 attr.ia_valid = ATTR_MODE;
1607 open_flags = nd->intent.open.flags;
1609 ctx = create_nfs_open_context(dentry, open_flags);
1610 error = PTR_ERR(ctx);
1614 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, ctx);
1618 error = nfs_intent_set_file(nd, ctx);
1622 put_nfs_open_context(ctx);
1626 put_nfs_open_context(ctx);
1633 #endif /* CONFIG_NFSV4 */
1636 * Code common to create, mkdir, and mknod.
1638 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1639 struct nfs_fattr *fattr)
1641 struct dentry *parent = dget_parent(dentry);
1642 struct inode *dir = parent->d_inode;
1643 struct inode *inode;
1644 int error = -EACCES;
1648 /* We may have been initialized further down */
1649 if (dentry->d_inode)
1651 if (fhandle->size == 0) {
1652 error = NFS_PROTO(dir)->lookup(NFS_SERVER(dir)->client, dir, &dentry->d_name, fhandle, fattr);
1656 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1657 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1658 struct nfs_server *server = NFS_SB(dentry->d_sb);
1659 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1663 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1664 error = PTR_ERR(inode);
1667 d_add(dentry, inode);
1672 nfs_mark_for_revalidate(dir);
1678 * Following a failed create operation, we drop the dentry rather
1679 * than retain a negative dentry. This avoids a problem in the event
1680 * that the operation succeeded on the server, but an error in the
1681 * reply path made it appear to have failed.
1683 static int nfs_create(struct inode *dir, struct dentry *dentry,
1684 umode_t mode, struct nameidata *nd)
1688 int open_flags = O_CREAT|O_EXCL;
1690 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1691 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1693 attr.ia_mode = mode;
1694 attr.ia_valid = ATTR_MODE;
1697 open_flags = nd->intent.open.flags;
1699 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, NULL);
1709 * See comments for nfs_proc_create regarding failed operations.
1712 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1717 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1718 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1720 if (!new_valid_dev(rdev))
1723 attr.ia_mode = mode;
1724 attr.ia_valid = ATTR_MODE;
1726 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1736 * See comments for nfs_proc_create regarding failed operations.
1738 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1743 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1744 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1746 attr.ia_valid = ATTR_MODE;
1747 attr.ia_mode = mode | S_IFDIR;
1749 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1758 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1760 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1764 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1768 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1769 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1771 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1772 /* Ensure the VFS deletes this inode */
1773 if (error == 0 && dentry->d_inode != NULL)
1774 clear_nlink(dentry->d_inode);
1775 else if (error == -ENOENT)
1776 nfs_dentry_handle_enoent(dentry);
1782 * Remove a file after making sure there are no pending writes,
1783 * and after checking that the file has only one user.
1785 * We invalidate the attribute cache and free the inode prior to the operation
1786 * to avoid possible races if the server reuses the inode.
1788 static int nfs_safe_remove(struct dentry *dentry)
1790 struct inode *dir = dentry->d_parent->d_inode;
1791 struct inode *inode = dentry->d_inode;
1794 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1795 dentry->d_parent->d_name.name, dentry->d_name.name);
1797 /* If the dentry was sillyrenamed, we simply call d_delete() */
1798 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1803 if (inode != NULL) {
1804 nfs_inode_return_delegation(inode);
1805 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1806 /* The VFS may want to delete this inode */
1808 nfs_drop_nlink(inode);
1809 nfs_mark_for_revalidate(inode);
1811 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1812 if (error == -ENOENT)
1813 nfs_dentry_handle_enoent(dentry);
1818 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1819 * belongs to an active ".nfs..." file and we return -EBUSY.
1821 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1823 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1826 int need_rehash = 0;
1828 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1829 dir->i_ino, dentry->d_name.name);
1831 spin_lock(&dentry->d_lock);
1832 if (dentry->d_count > 1) {
1833 spin_unlock(&dentry->d_lock);
1834 /* Start asynchronous writeout of the inode */
1835 write_inode_now(dentry->d_inode, 0);
1836 error = nfs_sillyrename(dir, dentry);
1839 if (!d_unhashed(dentry)) {
1843 spin_unlock(&dentry->d_lock);
1844 error = nfs_safe_remove(dentry);
1845 if (!error || error == -ENOENT) {
1846 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1847 } else if (need_rehash)
1853 * To create a symbolic link, most file systems instantiate a new inode,
1854 * add a page to it containing the path, then write it out to the disk
1855 * using prepare_write/commit_write.
1857 * Unfortunately the NFS client can't create the in-core inode first
1858 * because it needs a file handle to create an in-core inode (see
1859 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1860 * symlink request has completed on the server.
1862 * So instead we allocate a raw page, copy the symname into it, then do
1863 * the SYMLINK request with the page as the buffer. If it succeeds, we
1864 * now have a new file handle and can instantiate an in-core NFS inode
1865 * and move the raw page into its mapping.
1867 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1869 struct pagevec lru_pvec;
1873 unsigned int pathlen = strlen(symname);
1876 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1877 dir->i_ino, dentry->d_name.name, symname);
1879 if (pathlen > PAGE_SIZE)
1880 return -ENAMETOOLONG;
1882 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1883 attr.ia_valid = ATTR_MODE;
1885 page = alloc_page(GFP_HIGHUSER);
1889 kaddr = kmap_atomic(page);
1890 memcpy(kaddr, symname, pathlen);
1891 if (pathlen < PAGE_SIZE)
1892 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1893 kunmap_atomic(kaddr);
1895 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1897 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1898 dir->i_sb->s_id, dir->i_ino,
1899 dentry->d_name.name, symname, error);
1906 * No big deal if we can't add this page to the page cache here.
1907 * READLINK will get the missing page from the server if needed.
1909 pagevec_init(&lru_pvec, 0);
1910 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1912 pagevec_add(&lru_pvec, page);
1913 pagevec_lru_add_file(&lru_pvec);
1914 SetPageUptodate(page);
1923 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1925 struct inode *inode = old_dentry->d_inode;
1928 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1929 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1930 dentry->d_parent->d_name.name, dentry->d_name.name);
1932 nfs_inode_return_delegation(inode);
1935 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1938 d_add(dentry, inode);
1945 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1946 * different file handle for the same inode after a rename (e.g. when
1947 * moving to a different directory). A fail-safe method to do so would
1948 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1949 * rename the old file using the sillyrename stuff. This way, the original
1950 * file in old_dir will go away when the last process iput()s the inode.
1954 * It actually works quite well. One needs to have the possibility for
1955 * at least one ".nfs..." file in each directory the file ever gets
1956 * moved or linked to which happens automagically with the new
1957 * implementation that only depends on the dcache stuff instead of
1958 * using the inode layer
1960 * Unfortunately, things are a little more complicated than indicated
1961 * above. For a cross-directory move, we want to make sure we can get
1962 * rid of the old inode after the operation. This means there must be
1963 * no pending writes (if it's a file), and the use count must be 1.
1964 * If these conditions are met, we can drop the dentries before doing
1967 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1968 struct inode *new_dir, struct dentry *new_dentry)
1970 struct inode *old_inode = old_dentry->d_inode;
1971 struct inode *new_inode = new_dentry->d_inode;
1972 struct dentry *dentry = NULL, *rehash = NULL;
1975 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1976 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1977 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1978 new_dentry->d_count);
1981 * For non-directories, check whether the target is busy and if so,
1982 * make a copy of the dentry and then do a silly-rename. If the
1983 * silly-rename succeeds, the copied dentry is hashed and becomes
1986 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1988 * To prevent any new references to the target during the
1989 * rename, we unhash the dentry in advance.
1991 if (!d_unhashed(new_dentry)) {
1993 rehash = new_dentry;
1996 if (new_dentry->d_count > 2) {
1999 /* copy the target dentry's name */
2000 dentry = d_alloc(new_dentry->d_parent,
2001 &new_dentry->d_name);
2005 /* silly-rename the existing target ... */
2006 err = nfs_sillyrename(new_dir, new_dentry);
2010 new_dentry = dentry;
2016 nfs_inode_return_delegation(old_inode);
2017 if (new_inode != NULL)
2018 nfs_inode_return_delegation(new_inode);
2020 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
2021 new_dir, &new_dentry->d_name);
2022 nfs_mark_for_revalidate(old_inode);
2027 if (new_inode != NULL)
2028 nfs_drop_nlink(new_inode);
2029 d_move(old_dentry, new_dentry);
2030 nfs_set_verifier(new_dentry,
2031 nfs_save_change_attribute(new_dir));
2032 } else if (error == -ENOENT)
2033 nfs_dentry_handle_enoent(old_dentry);
2035 /* new dentry created? */
2041 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2042 static LIST_HEAD(nfs_access_lru_list);
2043 static atomic_long_t nfs_access_nr_entries;
2045 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2047 put_rpccred(entry->cred);
2049 smp_mb__before_atomic_dec();
2050 atomic_long_dec(&nfs_access_nr_entries);
2051 smp_mb__after_atomic_dec();
2054 static void nfs_access_free_list(struct list_head *head)
2056 struct nfs_access_entry *cache;
2058 while (!list_empty(head)) {
2059 cache = list_entry(head->next, struct nfs_access_entry, lru);
2060 list_del(&cache->lru);
2061 nfs_access_free_entry(cache);
2065 int nfs_access_cache_shrinker(struct shrinker *shrink,
2066 struct shrink_control *sc)
2069 struct nfs_inode *nfsi, *next;
2070 struct nfs_access_entry *cache;
2071 int nr_to_scan = sc->nr_to_scan;
2072 gfp_t gfp_mask = sc->gfp_mask;
2074 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2075 return (nr_to_scan == 0) ? 0 : -1;
2077 spin_lock(&nfs_access_lru_lock);
2078 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2079 struct inode *inode;
2081 if (nr_to_scan-- == 0)
2083 inode = &nfsi->vfs_inode;
2084 spin_lock(&inode->i_lock);
2085 if (list_empty(&nfsi->access_cache_entry_lru))
2086 goto remove_lru_entry;
2087 cache = list_entry(nfsi->access_cache_entry_lru.next,
2088 struct nfs_access_entry, lru);
2089 list_move(&cache->lru, &head);
2090 rb_erase(&cache->rb_node, &nfsi->access_cache);
2091 if (!list_empty(&nfsi->access_cache_entry_lru))
2092 list_move_tail(&nfsi->access_cache_inode_lru,
2093 &nfs_access_lru_list);
2096 list_del_init(&nfsi->access_cache_inode_lru);
2097 smp_mb__before_clear_bit();
2098 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2099 smp_mb__after_clear_bit();
2101 spin_unlock(&inode->i_lock);
2103 spin_unlock(&nfs_access_lru_lock);
2104 nfs_access_free_list(&head);
2105 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
2108 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2110 struct rb_root *root_node = &nfsi->access_cache;
2112 struct nfs_access_entry *entry;
2114 /* Unhook entries from the cache */
2115 while ((n = rb_first(root_node)) != NULL) {
2116 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2117 rb_erase(n, root_node);
2118 list_move(&entry->lru, head);
2120 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2123 void nfs_access_zap_cache(struct inode *inode)
2127 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2129 /* Remove from global LRU init */
2130 spin_lock(&nfs_access_lru_lock);
2131 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2132 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2134 spin_lock(&inode->i_lock);
2135 __nfs_access_zap_cache(NFS_I(inode), &head);
2136 spin_unlock(&inode->i_lock);
2137 spin_unlock(&nfs_access_lru_lock);
2138 nfs_access_free_list(&head);
2141 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2143 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2144 struct nfs_access_entry *entry;
2147 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2149 if (cred < entry->cred)
2151 else if (cred > entry->cred)
2159 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2161 struct nfs_inode *nfsi = NFS_I(inode);
2162 struct nfs_access_entry *cache;
2165 spin_lock(&inode->i_lock);
2166 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2168 cache = nfs_access_search_rbtree(inode, cred);
2171 if (!nfs_have_delegated_attributes(inode) &&
2172 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2174 res->jiffies = cache->jiffies;
2175 res->cred = cache->cred;
2176 res->mask = cache->mask;
2177 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2180 spin_unlock(&inode->i_lock);
2183 rb_erase(&cache->rb_node, &nfsi->access_cache);
2184 list_del(&cache->lru);
2185 spin_unlock(&inode->i_lock);
2186 nfs_access_free_entry(cache);
2189 spin_unlock(&inode->i_lock);
2190 nfs_access_zap_cache(inode);
2194 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2196 struct nfs_inode *nfsi = NFS_I(inode);
2197 struct rb_root *root_node = &nfsi->access_cache;
2198 struct rb_node **p = &root_node->rb_node;
2199 struct rb_node *parent = NULL;
2200 struct nfs_access_entry *entry;
2202 spin_lock(&inode->i_lock);
2203 while (*p != NULL) {
2205 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2207 if (set->cred < entry->cred)
2208 p = &parent->rb_left;
2209 else if (set->cred > entry->cred)
2210 p = &parent->rb_right;
2214 rb_link_node(&set->rb_node, parent, p);
2215 rb_insert_color(&set->rb_node, root_node);
2216 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2217 spin_unlock(&inode->i_lock);
2220 rb_replace_node(parent, &set->rb_node, root_node);
2221 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2222 list_del(&entry->lru);
2223 spin_unlock(&inode->i_lock);
2224 nfs_access_free_entry(entry);
2227 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2229 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2232 RB_CLEAR_NODE(&cache->rb_node);
2233 cache->jiffies = set->jiffies;
2234 cache->cred = get_rpccred(set->cred);
2235 cache->mask = set->mask;
2237 nfs_access_add_rbtree(inode, cache);
2239 /* Update accounting */
2240 smp_mb__before_atomic_inc();
2241 atomic_long_inc(&nfs_access_nr_entries);
2242 smp_mb__after_atomic_inc();
2244 /* Add inode to global LRU list */
2245 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2246 spin_lock(&nfs_access_lru_lock);
2247 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2248 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2249 &nfs_access_lru_list);
2250 spin_unlock(&nfs_access_lru_lock);
2254 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2256 struct nfs_access_entry cache;
2259 status = nfs_access_get_cached(inode, cred, &cache);
2263 /* Be clever: ask server to check for all possible rights */
2264 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2266 cache.jiffies = jiffies;
2267 status = NFS_PROTO(inode)->access(inode, &cache);
2269 if (status == -ESTALE) {
2270 nfs_zap_caches(inode);
2271 if (!S_ISDIR(inode->i_mode))
2272 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2276 nfs_access_add_cache(inode, &cache);
2278 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2283 static int nfs_open_permission_mask(int openflags)
2287 if ((openflags & O_ACCMODE) != O_WRONLY)
2289 if ((openflags & O_ACCMODE) != O_RDONLY)
2291 if (openflags & __FMODE_EXEC)
2296 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2298 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2301 int nfs_permission(struct inode *inode, int mask)
2303 struct rpc_cred *cred;
2306 if (mask & MAY_NOT_BLOCK)
2309 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2311 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2313 /* Is this sys_access() ? */
2314 if (mask & (MAY_ACCESS | MAY_CHDIR))
2317 switch (inode->i_mode & S_IFMT) {
2321 /* NFSv4 has atomic_open... */
2322 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
2323 && (mask & MAY_OPEN)
2324 && !(mask & MAY_EXEC))
2329 * Optimize away all write operations, since the server
2330 * will check permissions when we perform the op.
2332 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2337 if (!NFS_PROTO(inode)->access)
2340 cred = rpc_lookup_cred();
2341 if (!IS_ERR(cred)) {
2342 res = nfs_do_access(inode, cred, mask);
2345 res = PTR_ERR(cred);
2347 if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2350 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2351 inode->i_sb->s_id, inode->i_ino, mask, res);
2354 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2356 res = generic_permission(inode, mask);
2362 * version-control: t
2363 * kept-new-versions: 5